Campbell R. Harvey

View this assignment, a real world exercise in top down global asset management. Your team will allocate into at least five global asset classes. You are free to choose the classes (they need not be the same as assignment 2). The classes can be equities or bonds. The objective of the analysis is to provide the inputs necessary to choose optimal portfolio weights.

I expect that you will provide me with a minimum of:

- Introduction to each market (with some brief background). A discussion of the risk of the market and the economic outlook. Using a common currency base for calculating returns, correlate with the MSCI world equity return or the U.S. return. Estimate correlations over different time periods and plot. Discuss the implications. Discuss the fundamental risks of each market. [Do not just copy some content from some Internet site. I want your analysis.]
**You will need to collect some predictive variables. Part of the assignment is finding interesting data.**. State clearly the*economic reasons*why you are choosing each variable that will be used in the prediction of the asset return. Please note that if the variable does not work, that is interesting too.- Summary of predictive regressions (your best model for each asset
class). Use monthly data and common base currency returns.
Try to get as much of a history as is possible [data from 1970 is available from
MSCI and the updated file resides in the data directory on the student drive].
Be sure to report diagnostics like Durbin-Watson statistics and
adjusted R
^{2}and t-ratios for each coefficient. Discuss the sign of each coefficient and whether it matches your intuition. Provide a correlation matrix of the**variables**in your model [*correlation of the variables not the correlation of the coefficients*]. I do not want any regression output dumped into the assignment. You can, however, provide an appendix summarizing the alternative models that you examined (but no output). - In-sample analysis. For each asset class, provide:

a. Raw correct direction count and percentage. [If you forecast a positive (negative) return and the realized return is positive (negative), that is a correct direction.]

b. Time-series plots of forecasts and actuals.

c. Elementary trading strategy analysis

Compare mean and standard deviation of buy and hold to a strategy of swapping out of equities into 30-day Eurocurrency deposit (in your common base currency available in Datastream) or the 30-day T-bill return (in the Ibbotson main file). That is, if forecasted asset class return is less than the 30-day rate take 100% cash position. If forecasted asset class return is greater than the 30-day rate, then go 100% in asset class. Be careful to get the units correct. The Eurocurrency rates are usually quoted on an annual basis. Either convert them to monthly or convert your asset class forecast to annual terms. Rerun the strategy with a filter rule. Only go long equity if forecasted returns exceeds Eurorate by 1.2% (annualized). If this filter is too big, you can try a smaller one, like 0.6%. - Out-of-sample prediction. Use coefficients estimated up to 2005:12 to form forecasts for 2006:01. (If you are unable to get data through December, you can forecast another month.
- Save the residuals from your forecasting models. Calculate the variance-covariance matrix of the residuals. You will use this in the next step
- Run your asset allocation model with the conditional inputs (your forecasted returns, and the variance-covariance matrix of the residuals). Given your taste for variance (which might be, for example, the unconditional variance of the U.S. equity return), determine the out-of-sample allocation weights. Restrict short and long positions to min -20% and max 50%, respectively.
- For
**one**asset class, save the set of residuals from your final forecasting model. Square the residuals. Try to estimate a GARCH(1,1) model with garch.xls. You will paste your residuals into column C. It is difficult to get this to run in Excel. Don't waste a lot of time if it does not converge properly. *No calculations necessary for this question.*Based on our results, how might we execute a multivariate trading strategy? Would you use the results from (6)? What additional regressions must be run? One unsophisticated (but potentially profitable) version might be to go long the country with the highest forecasted return. Think about a more sophisticated version. You do not have to execute the strategy -- just tell me how you would do it.*No calculations necessary for this question.*We have treated all returns in a common currency base. How do you interpret the optimization in the common currency? Hedged or unhedged? What does this imply about our desire to hedge the currency exposure? Does it make sense to optimize in local currency terms? Comment in words about the various alternatives that we might consider to handle currency exposure.